Something to look forward to: Rumors and leaks regarding Nvidia's next-generation graphics chips continue to flow as the company's plans likely remain in flux. Although the RTX 50 series flagship product will almost certainly launch this year, the status of its companions remains unclear. Meanwhile, the lineup of AI chips that follows is already taking shape.
Established leaker @kopite7kimi recently claimed that Nvidia might begin shipping the RTX 5080 before the more powerful 5090. New GPU generations typically debut with the flagship product, so this strategy would be somewhat unusual.
The projected release dates of the upcoming RTX 50 series graphics cards, codenamed Blackwell, have recently fluctuated between late 2024 and early 2025. Most predictions place the 5090 and 5080 before the end of this year, with mid-range and mainstream variants following in 2025. However, YouTuber Moore's Law is Dead updated his information to suggest that only the top-end GPU would launched in 2024, prompting Kopite to respond, stating that the 5080 should debut first.
It's not ture. RTX 5080 should be released first.
– kopite7kimi (@kopite7kimi) May 7, 2024
One potential reason for the switch-up is that the RTX 5080 might outperform the most powerful Nvidia GPU currently available – the 4090 – thus making it the most potent graphics card on the market until the 5090 begins shipping. Most reports indicate that rival vendor AMD will introduce the RDNA 4 lineup this year, but it will only consist of mid-range products. However, sources recently informed Benchlife that the new Radeon series could be delayed until 2025. Intel plans to release its first-ever enthusiast-class GPU in 2024, but there is still no indication of how it compares to Blackwell.
Benchlife also reports that coolers for the RTX 50 series are currently undergoing testing, revealing four designs ranging between 250w and 600w. Trial wattage ratings don't necessarily reflect final products, so customers shouldn't expect the 5090 to require anywhere near 600w.
For instance, early reports surrounding the 40 series sparked concern that its flagship might draw over 800W, but the 4090 ended up only using 450W. On the opposite end of the scale, the preliminary 250W tests for Blackwell could represent an RTX 5070 or 5070 Ti.
Looking further ahead, analyst Ming-Chi Kuo reports that the successor to Nvidia's Blackwell AI GPUs, codenamed "Rubin," will begin mass production in the fourth quarter of 2025, with rack solutions following in early 2026.
The lineup's inaugural chip, the R100, will advance to TSMC's 3nm N3 node, continue using CoWoS-L packaging, incorporate a 4x reticle design, and feature HBM4 memory. The GR200 Grace CPU will also utilize N3, moving the entire system beyond Blackwell's 5nm node.